Research On 3D Reconstruction Of Large Structure Based On Image Sequences

Nowadays,more and more aspects need 3D reconstruction technology gradually,such as digital display of buildings,the protection or display of antiquities,topographic survey and industrial survey.Because of the low cost and easy operation,3D reconstruction technology based on image sequences is often used to obtain 3D models of objects.When reconstructing the large-scale structure,UAV always be chosen as a platform flying around the object to acquire images.And then the camera's moving position and the corresponding 3D model of the scene can be reconstructed through these captured images.In order to achieving the purpose of reconstructing,displaying and measuring the 3D reconstruction of large structure,this paper set up a system to reconstruct and measure large structure,also optimize the relevant algorithm.The main work and achievements are as follows:(1)An image acquisiton platform based on UAV is set up through hardware selection.And the calibration of this system is completed,so that it can be used to capture object images through the UAV flying.(2)The methods of estimating the camera motion and registering the sparse 3D points by image sequences is researched.The joint keypoints extracting algorithm is used to increase the extreacting number.Also the hash method and pre-matching method are used to speed up this matching process.At the same time,the explicit unique points such as salient background is used to re-correct the reconstruction result of the duplicate or symmetric structure.(3)A method in dense 3D points reconstruction process based on multi-scale images is proposed.The finally reconstruction of dense point cloud is combined of the result from low resolution and high resolution,which can not only obtain good results but also improve the efficiency in some degree.And then,the improved isosurface extraction method is used to mesh and construct the surface in which space is divided by octree method.(4)In order to retrieve the missing scale of 3D objects,as the reconstruction model is up-to-scale,a scale recovery method based on image sequence is proposed.The local point cloud obtained by binocular camera is used to restore the whole model's actual size and then scale it,so that the updating model can be used to measure real size.(5)The software of 3D reconstruction and measurement system based on image sequences is developed.Based on the open source library,the whole 3D reconstruction process and dimension measurement function are realized.